iupui scholarworks repository
TRANSCRIPT
))) ) ) ) )) ) ) ))) )) ) )))
� � � �� � �
nnnnnnnnnnn
nnnnnnnnnnnnnn n n nn n nn n nnnn n nnnn n n n n n n n n n nn n nnnn n
n n n n nnn nnnnn n n n n nn g n n nnnnnn nnn
innninnin
n nnnnnnnn
nn nmn
innninnin
g n nnmnnn 333 33 333
PERFORMANCE EVALUATION OF ROUTING PROTOCOLS USING NS-2
AND REALISTIC TRACES ON DRIVING SIMULATOR
A Thesis
Submitted to the Faculty
of
Purdue University
by
Mingye Chen
In Partial Fulfillment of the
Requirements for the Degree
of
Master of Science in Electrical and Computer Engineering
August 2013
Purdue University
Indianapolis, Indiana
ii
ACKNOWLEDGMENTS
I would like to express my sincere gratitude to my adviser, Professor Lingxi Li,
for his excellent unconditional support throughout the entire course of my graduate
study.
I also want to extend my thanks to the faculty at IUPUI for their help and
contributions to my thesis work. My committee members, Professor Yaobin Chen
and Professor Dongsoo Kim, who generously shared their time and experience with
me. I would like to thank them for their efforts on discussing the research problems
with me. I also want to express my thanks to Professor Brian King and Dr. Renran
Tian for their guidance and suggestions during my study.
I also would like to acknowledge my family members and friends for their support.
iii
TABLE OF CONTENTS
Page
LIST OF TABLES . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . v
LIST OF FIGURES . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . vi
ABBREVIATIONS . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . vii
ABSTRACT . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . ix
1 INTRODUCTION . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1
1.1 Background . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1
1.2 Why VANET . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3
1.3 VANET applications . . . . . . . . . . . . . . . . . . . . . . . . . . 4
1.3.1 Overview . . . . . . . . . . . . . . . . . . . . . . . . . . . . 4
1.3.2 Periodic messages . . . . . . . . . . . . . . . . . . . . . . . . 5
1.3.3 Event-driven messages . . . . . . . . . . . . . . . . . . . . . 5
1.3.4 Safety-related applications . . . . . . . . . . . . . . . . . . . 6
1.4 Thesis contributions . . . . . . . . . . . . . . . . . . . . . . . . . . 9
2 ROUTING AND BROADCASTING . . . . . . . . . . . . . . . . . . . . 10
2.1 Multi-hop communication . . . . . . . . . . . . . . . . . . . . . . . 10
2.2 Proactive Routing . . . . . . . . . . . . . . . . . . . . . . . . . . . . 11
2.2.1 OLSR . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 11
2.3 Reactive routing . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 13
2.3.1 DSR . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 13
2.3.2 AODV . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 14
3 ISSUES IN VANET SIMULATION . . . . . . . . . . . . . . . . . . . . . 15
3.1 Mobility Models . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 15
3.2 Impact of Mobility Models . . . . . . . . . . . . . . . . . . . . . . . 17
3.3 Classifications of Mobility Models . . . . . . . . . . . . . . . . . . . 19
iv
Page
3.4 Artificial traces . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 21
4 SIMULATOR COMPARISON . . . . . . . . . . . . . . . . . . . . . . . . 22
4.1 VANET simulator overview . . . . . . . . . . . . . . . . . . . . . . 22
4.2 SUMO-based simulator . . . . . . . . . . . . . . . . . . . . . . . . . 23
4.2.1 MOVE . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 23
4.2.2 TraNS . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 25
4.3 NCTUns . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 25
4.4 VanetMobiSim . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 26
4.5 Summary . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 26
5 SIMULATION SETUP . . . . . . . . . . . . . . . . . . . . . . . . . . . . 28
5.1 Introduction to driving simulator . . . . . . . . . . . . . . . . . . . 28
5.1.1 Comparison with other models . . . . . . . . . . . . . . . . . 28
5.1.2 Architecture . . . . . . . . . . . . . . . . . . . . . . . . . . . 34
5.2 Simulation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 37
6 RESULT ANALYSIS . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 40
6.1 Evaluation metrics . . . . . . . . . . . . . . . . . . . . . . . . . . . 40
6.2 Comparison with RWP model . . . . . . . . . . . . . . . . . . . . . 40
6.3 Analysis of throughput at intersections . . . . . . . . . . . . . . . . 42
7 CONCLUSION AND FUTURE WORK . . . . . . . . . . . . . . . . . . 46
LIST OF REFERENCES . . . . . . . . . . . . . . . . . . . . . . . . . . . . 48
v
LIST OF TABLES
Table Page
1.1 Safety-related applications . . . . . . . . . . . . . . . . . . . . . . . . . 7
4.1 VANET simulator features . . . . . . . . . . . . . . . . . . . . . . . . . 27
5.1 Wireless simulation parameters . . . . . . . . . . . . . . . . . . . . . . 38
5.2 Traffic scenario parameters . . . . . . . . . . . . . . . . . . . . . . . . . 38
vi
LIST OF FIGURES
Figure Page
1.1 Vehicular network components . . . . . . . . . . . . . . . . . . . . . . . 3
1.2 An example of EEBL warning when Vehicle A is braking hard. [29] . . 6
2.1 Illustration of MPR problem . . . . . . . . . . . . . . . . . . . . . . . . 12
4.1 The structure of some VANET simulators [43] . . . . . . . . . . . . . . 22
4.2 The structure of SUMO [26] . . . . . . . . . . . . . . . . . . . . . . . . 24
5.1 Traffic Density at Intersection . . . . . . . . . . . . . . . . . . . . . . . 29
5.2 Vehicular density [11]: (a) IDM model (b) IDM-IM traffic light model . 29
5.3 Speed variation of 4 vehicles in a row appoaching red light . . . . . . . 31
5.4 Speed variation of vehicles approaching intersections in other models . 32
5.5 Traffic light signal transition over time . . . . . . . . . . . . . . . . . . 33
5.6 Evolution of vehicle speed at an intersection . . . . . . . . . . . . . . . 33
5.7 Architecture of proposed simulation platform . . . . . . . . . . . . . . 35
5.8 Map editor of HyperDrive . . . . . . . . . . . . . . . . . . . . . . . . . 36
5.9 Visualization of traffic scenarios . . . . . . . . . . . . . . . . . . . . . . 36
6.1 Packet Delivery Ratio with varying number of CBR . . . . . . . . . . . 41
6.2 Average Packet Delay with varying number of CBR . . . . . . . . . . . 41
6.3 Comparison of Driving Simulator traces and Random Waypoint . . . . 43
6.4 Average throughput transferred at Node 5 . . . . . . . . . . . . . . . . 44
6.5 Average throughput transferred at Node 71 . . . . . . . . . . . . . . . 45
vii
ABBREVIATIONS
VANET Vehicular Ad hoc Networks
MANET Mobile Ad hoc Networks
DSRC Dedicated Short Range Communication
WAVE Wireless Access in Vehicular Environment
NHTSA National Highway Traffic Safety Administration
U.S.DOT U.S. Department of Transportation
AAA Association of America
ITS Intelligent Transportation Systems
EEBL Emergency electronic brake lights
RWP Random Way Point
TIGER Topologically Integrated Geographic Encoding and Referencing
GDF Geographic Data File
KM Krauss Model
GM General Motors Model
IDM Intelligent Driver Model
IDM-IM Intelligent Driver Model with Intersection Management
IDM-LC Intelligent Driver Model with Lane Change
CanuMobiSim Communication in Ad Hoc Networks for Ubiquitous Computing
NS2 Network Simulator 2
GlomoSim Global Mobile Information System Simulator
MOVE MObility model generator for VEhicular networks
SUMO Simulation of Urban MObility
TraCI Traffic Control Interface
TraNs Traffic and Network Simulator
viii
NCTUns National Chiao Tung University Network Simulator
PDR Packet Delivery Ratio
OLSR Optimized Link Sate Routing
DSR Dynamic Source Routing
AODV Ad hoc On-Demand Vector routing
ix
ABSTRACT
Chen, Mingye. MSECE, Purdue University, August 2013. Performance Evaluationof Routing Protocols Using NS-2 and Realistic Traces on Driving Simulator. MajorProfessor: Lingxi Li.
With the rapid growth in wireless mobile communication technology, Vehicular
Ad-hoc Network (VANET) has emerged as a promising method to effectively solve
transportation-related issues. So far, most of researches on VANETs have been con-
ducted with simulations as the real-world experiment is expensive. A core problem
affecting the fidelity of simulation is the mobility model employed. In this thesis, a
sophisticated traffic simulator capable of generating realistic vehicle traces is intro-
duced. Combined with network simulator NS-2, we used this tool to evaluate the
general performance of several routing protocols and studied the impact of intersec-
tions on simulation results. We show that static nodes near the intersection tend to
become more active in packet delivery with higher transferred throughput.
1
1. INTRODUCTION
1.1 Background
The evolution of wireless communication technology could be traced back to 19th
century to early 20th century, when several great pioneers in scientific area made
breakthrough in engineering science. Thirteen years after Hertz discovery in 1888,
Guglielmo Marconi successfully transmitted wireless signals across the Atlantic be-
tween Poldhu, Cornwall and St. Johns Newfoundland. This unprecedented achieve-
ment enlarged the wireless transmission range to 3500 km. Stepping into the 20th
century, wireless technology has become the most widely adopted and rapidly devel-
oped digital cellular standard. And currently, many applications derived from wireless
communication that could save life and energy is now applied in daily human life,
which has made wireless communications one of the hottest research topics.
With potential of providing high-speed data transmission and reducing cost, wire-
less communication technologies are the best candidates for a wide range of applica-
tions. Among them, a newly flourished one is Vehicular Ad hoc Networks (VANETs).
Ad hoc networks refer to the networks with the absence of central or pre-established
infrastructure.
Ad hoc Networks are collection of self-governing mobile nodes [24]. Vehicular Ad
hoc Networks is the technology of establishing a mobile network composed of vehicles
capable of exchanging information with each other, it stands for a special type of
Mobile ad hoc Networks (MANETs). Since the late 1990s, A great deal of efforts have
been made in the research area of MANETs, vehicle-to-vehicle (V2V), and vehicle-
to-infrastructure (V2I) communications. VANET is one of the most significant areas
for the improvement of Intelligent Transportation System (ITS).
2
In general, the vehicular networks consist of two types of mobile nodes: Vehicles
and roadside units. Vehicles equipped with On Board Units (OBU) are able to com-
municate with each other and Road Side Units (RSUs). RSUs are supposed to be
placed at area with high vehicle density. They also play indispensable roles in many
scenarios, such as providing Internet gateway. However vehicular networks should
not rely on RSUs, as a high number of RSUs will increase the cost of this technology
drastically.
The primary motivation for VANETs was to improve driving safety and pro-
vide optimized traffic management to avoid congestion by interchanging data in real
time. Also, this rapidly emerging technology showed great potentials for Internet ac-
cess, inter-vehicle gaming, fast-flourishing infotainment industry, and other network-
related applications. These features have made VANETs gain a great deal of foci
over the past decades. The importance and potential impact of VANETs have been
confirmed by the rapid proliferation of consortia involving car manufacturers, various
government agencies, and academia.
Each year there are several national and international projects worldwide by gov-
ernment agencies, industry, and university focusing on the improvement and exten-
sion of VANET technology. Examples include, among many others, Vehicle Safety
Communications Consortium, the Advanced Safety Vehicle Program, the Car2Car
Communication Consortium (Europe), Vehicle Infrastructure Integration Program
[27], and efforts made on the standardization such as ‘WAVE’(Wireless Access in
Vehicular Environment), IEEE 802.11p [22].
Automotive manufacturers such as BMW, Toyota, and Ford have also announced
the plan to equip their vehicles with more powerful computational capability. Chrysler
was among the first manufacturers to integrate access to Internet in some of its 2009
line of products. As this trend continues, the number of vehicles equipped with
communication devices and wireless network interfaces will go high dramatically in the
near future. These vehicles will be able to run network protocols that by exchanging
messages, serve for a safer, and more fluid traffic on the roads.
3
Fig. 1.1. Vehicular networks include two parts: Vehicles and Infrastructures
1.2 Why VANET
Traffic accidents and congestion on the roads nowadays are among the most severe
issues in people′s daily life. The safety-related issues kept threatening human lives
and traffic congestion consumes tons of resources and time, they are both problems
that take tolls on personal life and national property.
A report from the Automobile Association of America (AAA) in 2008 stated that
according to the Federal Highway Administration [48], the cost of traffic per person
in 2005 is $3.2 million and $68,170 for deaths and injuries respectively. AAA also
estimates that traffic crashes each year will cost $166.7 billion, including emergency,
medical services, property damage, and productivity loss.
In 2010, more than 30000 people were killed in traffic accidents and more than
2 million people were injured, according to the U.S. Department of Transportation
(U.S.DOT) [46]. In 2011, 32,367 people died in car crashes and 2,217,000 people
were injured in motor vehicle crashes. An estimation by The National Highway
Traffic Safety Administration (NHTSA) [47] claimed that there are about 10 million
unreported crashes each year.
4
Also, a new report from the Texas A&M Transportation Institute claimed that
American people spent up to 5.5 billion additional hours sitting in traffic in 2011, and
2.9 billion gallons of gasoline were burned while sitting in traffic congestion.
In order to improve driving safety, there have been several generations of tech-
nologies, from passive precautions like seatbelts and airbags that reduce the damage
caused by crash to active safety technologies aiming to prevent collisions before they
occur.
However passive methods are not so effective against accidents, since in most
cases airbags and seatbelts cannot protect drivers and passengers from being injured
or even killed in collisions. So the best solution is to give driver the ability to foresee
the unsafe situations before accidents take place, in particular, the active safety, using
systems with devices such as radar and camera to detect possible hazard and give
alarm to drivers or even take over to prevent incidents from happening. Nevertheless,
device like radar and camera can only detect collisions in a limited angle, thus possible
incidents could still remain undetected in situations like sharp curves or traffic signal
violation at intersections. To achieve full awareness of possible accidents, construction
of a network that enables vehicles to exchange information is in need.
1.3 VANET applications
1.3.1 Overview
Dedicated Short Range Communication (DSRC) [44] which works in the 5.9 GHz
band in the US is specifically designed for applications in vehicular wireless applica-
tion. The maximum signal range of DSRC is 1000 meters.
DSRC is mainly composed by two types of communications: Vehicle to Vehicle
(V2V) or Vehicle Ad-hoc Networks (VANETs) and Vehicle to Infrastructure (V2I).
V2V refers to the data exchange among vehicles for safety and other purposes while
the communication between vehicles and road-side infrastructures serves as an exten-
sion to V2V such as Internet Gateway. However, the V2I part requires long-period
5
hardware construction which costs tremendous expenses and time, as a result this
thesis primarily focused on the V2V communication.
Most applications in VANET are based on periodical message exchange, and
some safety-related applications also send critical information triggered by emergen-
cy. Thus the former is defined as periodical messages and the latter is defined as
event-driven messages.
1.3.2 Periodic messages
Depending on the purpose of certain types of applications, they will either require
data from sensors or from other vehicles. Hence, periodic messages are sent for
awareness of the vicinity and detection of dangerous situations. Each vehicle will
include its current speed, position and direction in this type of data exchange. By
collection of this information, potential collision or congestion could be evacuated
through calculation. Apart from this functionality, there is another type of periodic
messages in uni-cast routing known as the Hello messages which is exchanged with
regular cycle to detect communication neighbors in the aspect of routing.
Since VANET is a large-scale network, too frequent exchange of periodic messages
could lead to the “Broadcast storm problem ”[41] and hinder the reception of messages
containing critical information.
1.3.3 Event-driven messages
Event-driven messages are sent when emergency occurs. This type of messages
needs to be generated and disseminated extremely fast. As a result, event-driven
messages are usually assigned with high priority to ensure its correctness and quick-
ness.
6
1.3.4 Safety-related applications
Current active safety technology is mainly based on single vehicle. The restriction
of single-vehicle-based safety technology has been stated above (e.g. the limited
detection area). Using DSRC, the V2V-based applications could be categorized into
the following parts [29]:
1. Intersection collision avoidance
2. Public safety
3. Sign extension
4. Vehicle diagnostics and maintenance
5. Information from other vehicles
Table 1.1 describes some of the applications designed for vehicular network, in-
cluding their function and features. Fig. 1.2 is an illustration of one of the emergency-
triggered applications, Emergency Electronic Brake Lights warning.
Fig. 1.2. An example of EEBL warning when Vehicle A is braking hard. [29]
7
Table 1.1: Safety-related applications
Classifications Applications Type of
Transmis-
sion
Data exchanged
Intersection
collision
avoidance
Stop sign viola-
tion warning
Periodic Directionality, position
of the stopping location,
weather condition, road
and surface type near the
stop sign
Left turn assis-
tant
Periodic Traffic signal light status,
cycle, and direction; road
type
Intersection Col-
lision Warning
Periodic and intersection informa-
tion; vehicle position, veloc-
ity, and heading
Traffic signal vi-
olation warning
Periodic Traffic signal light status,
cycle, direction, position
of the traffic signal lo-
cation, weather condition,
road shape
Stop sign move-
ment assistance
Periodic Vehicle speed, location, and
direction
Public Safety
Approaching e-
mergency vehi-
cle warning
Event-
driven
Emergency vehicle speed,
location, lane position, and
intended routes
continued on next page
8
Table 1.1: continued
Classifications Applications Type of
Transmis-
sion
Data exchanged
Emergency vehi-
cle signal pre-
emption
Event-
driven
Emergency vehicle position,
lane information
SOS services Event-
driven
Location, vehicle informa-
tion, and time
Post-collision
warning
Event-
driven
vehicle information, colli-
sion location
Sign extension
Wrong way driv-
er warning
Periodic Position, heading, and
warning
Work zone warn-
ing
Periodic Distance to work zone and
reduced speed limits
Curve speed
warning
Periodic Curve location, curve speed
limits, bank, and road sur-
face type
Vehicle Diagnos-
tics and Mainte-
nance
Safety recall no-
tice
Event-
driven
Safety recall message
Information
from Other
Vehicles
Cooperative
forward collision
warning
Periodic Position, velocity, accelera-
tion, heading, and yaw-rate
Emergency elec-
tronic brake
lights
Event-
driven
Position, direction, speed,
and deceleration
continued on next page
9
Table 1.1: continued
Classifications Applications Type of
Transmis-
sion
Data exchanged
Lane change
warning
Periodic Lane position, speed, direc-
tion, and acceleration
Blind spot warn-
ing
Periodic Speed, position, direction,
acceleration, and warning
for detection of vehicle at
blind spot
1.4 Thesis contributions
The focus of this thesis is a systematic analysis of VANET simulations. By the
development of parser between the driving simulator and ns2, the improvement of
the mobility models used in VANET simulation is achieved. The main contributions
are:
1) The extension of functionality on the driving simulator. The data of driving
simulator outputs are well organized with network simulator so that its application
in terms of network simulation is made possible.
2) The construction of a simulation platform that is able to generate realistic
traffic flow and vehicle movement trace for accurate VANET simulations. By an
integral comparison with the state-of-art simulators, the advantage of the proposed
simulator is validated.
3) Through protocol testing based on the simulator, some featured effects have
been observed which are helpful for further protocol development. These effects
require a simulator that can generate highly detailed and accurate traffic trace, hence
the importance of a good traffic simulator in VANET simulation is revealed.
10
2. ROUTING AND BROADCASTING
2.1 Multi-hop communication
Unlike traditional wireless communication, one of the features of ad hoc mobile
networks is that nodes aim to send messages to receivers out of the radio range. In
order to achieve this function, before reaching the destination, the messages will first
go through some intermediate mobile nodes which serve as relays to help forward the
data to the intended receiver. And each intermediate node is called a hop, and this
type of communication in ad hoc network is defined as multi-hop communication.
Hence, when the destination is several hops away from the sender, certain algorithms
should be used to find a path to eventually deliver the message. And therefore
routing protocol development has become one of the most challenging issues in ad
hoc networks.
Same as MANETs, the multi-hop communication is a core component of VANETs
as well. As stated in the previous chapters, the applications of VANETs primarily
rely on the data exchange: periodic messages for regular environment detection and
event-driven messages for some urgent situations. Thus gathering and disseminating
information with the help of neighboring vehicles has become an indispensable enabler
for many vehicular applications.
As we know, VANET is a subclass of MANET, though their difference mainly lies
in the movement pattern and scalability, they still share lots of attribute in terms of
ad hot network. Thus, many proposed strategies for routing in MANET could be a
good starting point for VANET routing development. For this reason, we will briefly
introduce some of relevant routing protocols in MANET and their feasibility.
11
2.2 Proactive Routing
In proactive routing algorithms, nodes gradually learn the network topology by ex-
changing messages. Each node originally gathers information to confirm the existence
of its one-hop neighbors, and then through information from one-hop neighbors, it
continuously learns the neighbor of neighbors over and over again. In this way, routes
are established to any destinations in the network and stored in the routing tables.
2.2.1 OLSR
One of proactive routing strategies is Optimized Link State Routing (OLSR) pro-
tocol [20] designed by the IETF community. It extends the traditional link state
routing with reduction of link state update overhead in ad hoc networks with high
node density.
As in proactive routing, nodes sense their neighborhood by periodically exchanging
HELLO messages. After the local vicinity is learned and the link state with neighbors
are confirmed, the local information is broadcasted via periodic topology control (TC)
messages throughout the whole network. Based on these TC messages, each node then
stretches its paths to further receivers. Each message is also tagged with sequence
number to be distinct with stale information.
However, when the size of network becomes large, the increment of TC messages
could become a huge burden of the network. In connection with this issue, the OLSR
takes advantage the multi point relay (MPR) technique [33] aiming to reduce the
amount of message flooding. The MPR approach requires a node to have known
routes to at least 2-hop neighbors. Then, the node selects among its 1-hop neighbors
a subset of these neighbors with minimum number of nodes as relays, which still
covers the routes to the complete 2-hop neighbors as the same 1-hop neighborhood
does. This subset of 1-hop neighbors is defined as the MPR set. If a message is
intended to reach the 2-hop neighborhood, only those nodes in the MPR set by the
12
Fig. 2.1. Illustration of the MPR. The MPR set of the source ismarked as black nodes. The times of retransmission have been re-duced.
source are chosen to forward the message. Fig. 2.1 is an example of how MPR reduces
the retransmission.
Obviously, the smaller the size of the MPR sets, the less the times the TC messages
need to be forwarded. Unfortunately, the finding the minimal MPR set is known to
be a problem of NP-complete [34] complexity level. A greedy heuristic has been
proposed in [33] with source node S:
1. Define a set MPR ′(S) with all the 2-hop neighbors of node S as its elements.
2. If there exist some direct neighbors that the only access to some of the 2-hop
neighbors of node S, then include these 1-hop neighbors in the set MPR(S), and
correspondingly delete all the 2-hop neighbors covered by these elements in MPR(S).
3. Until MRP′(S) is an empty set, keep picking elements out of the 1-hop neighbors
of node S left from MPR(S). Always choose the 1-hop neighbors with connections to
13
the largest number of elements left in MPR′(S) first. In the case of ties, select the
1-hop neighbors with the greatest scale of neighborhood.
Although the OLSR did great improvement in terms of reducing TC messages
flooding using the MPRs, the VANETs′ nature of being highly mobile frequently
changes the network topology. As a result, the routing overhead issue caused by
TC messages still has not been fully resolved despite that the OLSR is one of most
successful proactive routing protocols.
2.3 Reactive routing
Reactive routing (also known as on-demand routing) is another mechanism. The
main difference from proactive routing is that routes are only created when there is
need to transmit messages to destination in reactive routing. Otherwise, the network
remains silent.
2.3.1 DSR
A representation for the reactive routing is the Dynamic Source Routing (DSR)
protocol [19] made up of two main processes: route discovery and routing mainte-
nance.
When a node needs to send data, it first checks if there already exists a path
stored previously. If no routes have been established before, it starts to broadcast
route request (RREQ) to nearby nodes. The RREQ messages contain the destination
information and are rebroadcasted to further nodes until the destination is found or
the maximum count is reached. Those nodes that help forward the RREQ messages
also add their own addresses in the route, thus a full path is recorded in the message in
this way upon the arrival of RREQ at the intended destination. Then the destination
node simply returns a route reply (RREP) using the path stored in the RREQ back
to the source. A route is set up after the source successfully receives the RREP. In
14
addition, the route discovery process can easily avoid loop of RREQ forwarding by
discarding all the messages that already contained the receivers address.
As a result of the frequent change of network topology, a route could be possibly
broken. As part of the route maintenance process, every node within an active route
will sense the link state with its 1-hop neighbors through the link layer such as the
IEEE802.11 acknowledgement (ACK) [35]. If a link disconnection has been detected,
the transmitter will inform all the nodes using this link through route error (RERR)
messages. In this case, nodes will use either use an alternative route or begin a new
route search.
2.3.2 AODV
Ad hoc On-Demand Distance Vector (AODV) routing protocol [18] is another
approach of reactive routing. The route discovery process is similar with that of
DSR. AODV makes use of sequence number to indicate the freshness of messages
thus avoid RREQ messages loop. Any node in the network will automatically discard
the RREQ with lower sequence number.
In general, the reactive routing protocols generates relatively low overhead, how-
ever the response time is high because route discovery will take some time before
message is sent.
15
3. ISSUES IN VANET SIMULATION
3.1 Mobility Models
Apparently, the most accurate and persuasive way to test VANET protocols per-
formance is the real-world experiment, but this evaluation implementation does have
some inevitable drawbacks:
1) It is neither easy nor cheap to have such a high number of vehicles in real test-
bed as well as huge amount of drivers required. Under most circumstances, real-world
evaluation is infeasible for its most probably unaffordable cost.
2) The performance of VANET protocols vary conspicuously as the road conditions
changes. It is though difficult to obtain test field of different characters in reality.
3) Test scenarios cannot be perfectly duplicated over times, thus it is extremely
hard to compare between the performances of different protocols in exactly the same
condition.
As a result, simulation programs become the only appropriate evaluation tool.
The Simulation for VANETs is primarily a combination of two parts: Network com-
munication model and Mobility Model.
The mobility model refers to the node movement pattern in the simulation. It is
the only new issue in VANET simulation that significantly differs from MANET. In
particular, these differences mainly lie in the following aspects:
1) nodes in VANETs move much faster and speed limit could vary under different
circumstances;
2) movement in VANETs is not completely random since it is limited by street
layout and also traffic rules. Therefore, in order to develop a more realistic environ-
ment for simulation, we need to take into consideration the specific limitations for
vehicle movement.
16
Mobility models used are usually considered from both macroscopic and micro-
scopic view: While from the macroscopic aspects of a mobility model, we intend
vehicular movement constraints such as street layout, road characterizations includ-
ing number of lanes per road, traffic signs established at intersections, and traffic flow
attributes involving vehicular density, average speed, etc. For microsopic features, we
instead account for individual driver behavior and its interaction with other entities
such as vehicle, obstacle and intersections. The microscopic description of mobility
models is a decisive factor of generating realistic vehicular traces. A realistic mobility
model should contain the following elements [29]:
• Accurate and realistic street layout : The road topology used in VANET simula-
tion should include roads that consist of various number of lanes and confine effective
area for vehicle movements.
• Intersections governed by traffic signs: Maps should contain intersections which
vehicles should be able to recognize and take corresponding reactions. For instance,
deceleration upon stop signs are desired and left turn vehicles should give right of
way to those going straight without the protected left turn arrows,
• Lane changing models : Drivers are not supposed to keep drive in the same
manner throughout the course of the entire trip. Instead, lane changing and speed
variation should be considered apart from single car-following model.
• Smooth speed variation: Since vehicle speed does not stop or start up abruptly,
proper deceleration and acceleration values should be accounted for a more natural
speed variation.
• Intelligent driving patterns : Drivers interact with their environments, not only
with respect to static obstacles, but also to dynamic obstacles, such as neighboring
cars and pedestrians.
• Human behaviors : The driver behaviors should be concerned with humans fac-
tors.
17
• Non-uniform distribution of vehicles : The distribution of vehicles in urban are
should not be considered as uniform, as the traffic density should vary in different
sectors.
• Different types of vehicles : Similar with the logic to concern multiple driver mod-
els, different types of vehicles present their own attribute in size, speed, acceleration,
etc.
• Effect of the implemented applications : If the goal of a simulation also includes
the part of performance evaluation for some particular applications, then the influ-
ence on the mobility pattern from applications should also be accounted. This is
especially necessary for routing management related applications as the preferred
paths are being changed in real time. This is usually accomplished by a bidirection-
al coupled simulator, which also allows the feedback from network simulator to the
traffic generator part.
3.2 Impact of Mobility Models
Mobility models describe the location of mobile nodes as a function of time in
simulation area. Mobility Models in MANETs simulations always consider synthetic
models that aim to describe mobility patterns for nodes using mathematic equations.
Although a mobility model at a more detailed level will definitely increase the sim-
ulation time, the author in [21] still argued that realistic traffic pattern is in need
continuously, as it could give a more precise evaluation of VANETs applications and
protocols. Indeed, lots papers present that mobility pattern could largely influence
the simulation result. From [1] [3] [4] [21] [22], we know the mobility patterns adopt-
ed play vital roles in VANETs simulations and directly change the performance of
routing protocols.
The performance of routing protocols is compared using different mobility models
in terms of delivery ratio, link duration and routing overhead in [1]. Literature [30]
showed us the problem of speed decay in which during the first period of simulation
18
time, the node speed could decrease to some long-term steady value, and this could
undermine the averaged results over time. And in [3] the author one step further
explained the problem and proposed the corresponding solutions to it. And this in
turn reflects that the very detailed parts of the node movement patterns in mobile
network simulation could affect the simulation result and conclusions to draw.
The relationship between the fidelity of a traffic model and the routing protocols
performance is uncertain and complex. This is primarily because, even with the
same initial start points and destinations, vehicles tend to move completely with
different patterns under different models, and some changes of patterns will improve
the simulation results while the rest could result in worse performance on the contrary.
In [4], author systematically studied the impact of a few different types of mo-
bility models on routing protocol performance using the following metrics: Average
link duration, delivery ratio (ratio of the number of packets delivered to the num-
ber of packets sent) and routing overhead (number of routing control packets sent).
According to an integral comparison of routing protocol performance under various
mobility models, the author concluded that the movement patterns of nodes in sim-
ulations do have influence on performance of routing protocols and there is no clear
winner in terms of the metrics used in this thesis. In addition, the author also admit-
ted that the test-suite also has some impact on the performance of routing protocols,
which in turn, reflects that although mobility models could make the simulation result
different, the relation between them is not easy to define.
Some earlier stochastic models cannot provide an appropriate representation for
real-world traffic wave, such as the popular Random Walk Model [16] (also known as
Brownian motion) that assumes nodes move continuously without restriction in open
area. Random Waypoint Model (RWM) is an extension of the Random Walk Model,
added with additional pauses between variations in heading and speed. However the
realism of these models in terms of geographical movement is far from being realistic.
These models suffer from aggregation of nodes in the central simulation area while
vehicle movements can barely be defined as random motion as it is restricted firstly by
19
street bound, and subsequently by traffic rules, individual driving behavior. Detailed
studies have been done on the comparison of VANET simulations based on RWM
and comparatively realistic traffic models in [5] [8] [9], all these works demonstrate
a significant difference of results. The RWM has been modified, for instance in [2]
[13] [14][15], to better meet VANETs simulation requirements, however, in these
models nodes move in velocity chosen independently from others behaviors, leaving
the necessary interactions among them ignored.
In order to tackle these drawbacks, many researchers have developed node move-
ment patterns able to better model realistic traffic. In [11], the author proposed
VANETMobiSim, a VANET simulator that integrates increasingly detailed levels of
models for urban traffic, namely the Stop Sign Model (SSM), the Probabilistic Traf-
fic Sign Model (PTSM), and the Traffic Light Model (TLM) where the vehicles will
stop for certain amount of time at intersections according to specific traffic signs.
Also the evaluation of these models shows the impact of node cluster at intersections
on the routing protocol performance that the longer vehicles stop at intersections,
the higher delivery ratio achieved. There are several literatures [6] [7] applied Social
Network Theory on mobility patterns and emphasized the importance of considering
interactive effects individual drivers have on each other.
3.3 Classifications of Mobility Models
In simulation of mobile networks, there are primarily two methods of generating
movement patterns: Trace and Synthetic Models [28]. Ideally, real-world trace, which
refers to position data generated by observation and collection of real-world system
movement patterns, is the most convincing data to be integrated in VANETs simu-
lations. Nevertheless, since the VANETs have not yet been fully deployed, so far it
is extremely difficult to gather real-world vehicle traces. Hence, lots of the work has
been devoting in the area of synthetic modeling.
20
Synthetic models are characterized by mathematical equations that try to capture
the movement of mobile nodes. The advantage of such model is obvious: Having a
mathematical model gives a formal description of a mobility model, and by changing
parameters in the equations used we can study the details how mobility pattern could
affect simulation results. There have been several works [1], [40] that systematically
studied and classified the mobility models used in MANETs and VANETs simulations
from different perspectives.
Camp et al [1] discussed several Synthetic Models in MANETs. These models
could be classified into the following two categories:
(1) Entity Mobility Models: The core of Entity Mobility Model is imitation of nat-
ural movement of a single mobile node in the network. Following are some examples
of Entity Mobility Models:
• Random Walk Model
• Random Waypoint Model
• Random Direction Model
• A Boundless Simulation Area Model
• Gauss-Markov Model
• A Probabilistic Version of the Random Walk Model
• City Section Model
(2) Group Mobility Models: Group Mobility Models try to mimic the interactions
among nodes in a group on the decision where to move. Following are some examples:
• Exponential Correlated Random Model
• Column Mobility Model
• Nomadic Community Mobility Model
• Pursue Mobility Model
• Reference Point Group Mobility Model
In [40], the synthetic models are divided into the 5 main types due to different
criteria of traffic patterns:
• Stochastic model: generates motion of entire randomness.
21
• Traffic Stream model: focuses on the mechanical properties of mobility model.
• Car Following model: vehicle to vehicle interaction within the same lane (speed
adjustment).
• Queue model: considers streets as buffers and vehicles enqueued in the buffers.
• Behavioral models: examines how movement is influenced by social interaction.
One of the deadly drawbacks of synthetic models is that the movement behaviors
generated are simply unnatural, which has been confirmed by examination of collected
real traces in [32].
3.4 Artificial traces
Though real-world vehicle traces are not available in most cases, lots of traffic sim-
ulators have been developed to generate artificial vehicle movements. The simulators
able to provide vehicle traces include commercial tools such as TSIS-CORSIM (Traffic
Software Integrated SystemCCorridor Simulation) [23], VISSIM [24], and PARAM-
ICS [25] and publicly available open-source tool like VanetMobiSim [5], and SUMO
(Simulation of Urban Mobility) [26] [27]. These tools could generate comparatively
realistic traces based on real-world map from resources such as TIGER (Topologically
Integrated Geographic Encoding and Referencing system) database [31], while issues
are that some of simulators are not freely accessible and those free tools developed
by universities or groups either have limited scenario selection or fidelity in traffic
generation. A more detailed comparison will be presented in later chapters.
22
4. SIMULATOR COMPARISON
4.1 VANET simulator overview
The simulation of VANETs is essentially different from that of MANETs, because
apart from the network modeling, it also needs to account for lots of elements of traffic
pattern at a detailed level: movement constrained by street layout, interactive driving
behavior, and intersection management. Therefore a complete VANET simulator
should at least consist of two main components: a network simulator and a traffic
simulator. However, traditional MANET simulations usually are more leaning to
consider network part and use simple mobility models. Therefore current VANET
simulation research has been focusing on development of traffic simulators that could
closely imitate real-world traffic flows and the integration with network simulators.
Fig. 4.1 listed some VANET simulator developed universities and their structure.
This section will introduce some of the freely available simulators and their features
and weaknesses by looking into the network and traffic simulator separately. Fig. 4.1
listed some VANET simulator developed universities and their structure.
Fig. 4.1. The structure of some VANET simulators [43]
23
Currently there is no standard regarding what properties simulators should pos-
sess to accurately simulate mobile networks under vehicular environments. But in
general, as many researchers have been stepping up their effort, a good VANETs
simulator should be able to reflect real traffic scenarios as much as possible. There
already exist a few VANET simulators, while they can rarely provide a proper simula-
tion environment. Some of traffic simulators are powerful such as the TSIS-CORSIM,
VISSIM [24], and PARAMICS [25], but they are not publicly available with a single
license of above 9000 USD, plus there is no effort taken to couple them with net-
work simulators. And other traffic simulators suffer from various problems such as
unnatural driving behavior and limited interaction with objects and intersections.
4.2 SUMO-based simulator
SUMO (Simulation of Urban Mobility) [26][27] is a widely used open-source mo-
bility generator written in C++ based on the Random Waypoint and the Krauc¯ar-
following model. The advantage of this simulator is its integration with TIGER
database to generate real-world-based topology. SUMO supports output files with
multiple forms to be compatible with various network simulators such as NS-2 [10].
Its input files are nodes.xml with the following information: node id, X coordinate
and Y coordinate. The other input file is edges.xml which defines road topology in-
cluding origin node, destination node, edge ID and number of lanes. Fig. 4.2 gives a
detailed description of SUMO structure.
4.2.1 MOVE
MOVE (Mobility model generator for Vehicular networks) [36] is an extension to
SUMO based on JAVA. The main contribution of MOVE is providing user-friendly
interface that allows user to finish all the configuration procedure by several mouse
clicking. MOVE also adds function of Google Earth maps that allows user to define
topology with real world information.
24
Fig. 4.2. The structure of SUMO [26]
25
While generating vehicle traces, MOVE accounts for some mircoscopic-level fac-
tors such as car-following. However lane-changing and intersection management is
completely ignored in this application.
4.2.2 TraNS
TraNS (Traffic and Network Simulation Environment) [36] is a VANET simulator
written in JAVA. It is the first effort that combines the network simulator (NS-2) and
the traffic simulator (SUMO). TraNS translates the output from SUMO into NS-2
readable form.
TraNS provides two modes: network-centric and application-centric mode. The
former does not have bidirectional feedback between network simulator and traf-
fic simulator while the later does. The application-centric mode synchronizes using
TraCI (Traffic Control Interface) [37]. MOVE, TraNS and some other SUMO based
simulator have made some effort in terms of generating realistic data for VANET sim-
ulation from various perspectives. However, since SUMO can only support realistic
road topology but lack the ability to simulate naturalistic driving behavior, hence the
core issue of VANET simulation still remains unresolved in this series of simulators.
4.3 NCTUns
NCTUns (National Chiao Tung University Network Simulator) [38] is a tightly
coupled VANET simulator which allows bidirectional communication between net-
work simulator and traffic simulator. Its first version NCTUns 1.0 was a mere network
simulator.
The vehicle movement pattern in NCTUns considers different types of road and
vehicle parameters like initial speed, desired speed, initial and maximum accelera-
tion/deceleration, and so on. The main drawback is the network part of this simula-
tor is not validated and the code for vehicle movement is tightly integrated with the
network simulation code which makes it difficult to extend.
26
4.4 VanetMobiSim
VanetMobiSim is an extension to CanuMobiSim [39]. The CanuMobisim is orig-
inally designed for MANET simulation use, it is unable to produce high levels of
realistic vehicular scenarios as it is mainly based on stochastic mobility models.
VanetMobisim extends CanuMobisim with implementations of IDM-IM (IDM
with Intersection Management) and IDM-LC (IDM with Lane Changes) mobility
models such that vehicles will behave accordingly in connection with intersections
and other drivers. VanetMobisim is one of the most advanced simulators compared
with other freely available ones since it has been validated against some commercial
traffic generator.
4.5 Summary
This chapter introduces the state-of-art of simulator in VANET research. Though
these tools pave the way from MANET to VANET simulation, they still have many
weakness hence cannot provide a completely suitable environment for VANET simu-
lation. Table 4.1 gives a summary of features of these simulators.
27
Table 4.1VANET simulator features
Features SUMO/MOVE/TraNs VanetMobisim NCTuns
Custom
Graphs
Supported Supported Supported
Random
Graphs
Grid Based Clustered
Voronoi Graphs
SHAPE-
File
Graphs
from Maps
Topologically Inte-
grated Geographic
Encoding and Ref-
erencing system
(TIGER) database
Geographic Da-
ta File (GDF)
Bitmap
image
Multilane
Graphs
Supported Supported Supported
Start/End
location
Random Random Random
Path Random Walk Random Walk Random
Walk
Velocity Road Dependent Road Dependent Road De-
pendent
Driving
Patterns
Car following Model Car Following Models, In-
telligent driver model, ex-
tended with Lane Changes
(IDM-LC) and Intersection
Management (IDM-IM)
Lane
changing
Not supported MOBIL Supported
28
5. SIMULATION SETUP
5.1 Introduction to driving simulator
5.1.1 Comparison with other models
For macro-traffic model, we usually consider the type of roads (one-way or bi-
directional), number of lanes, environment (urban, suburban, freeway, etc.), traffic
density, and speed limits. However, our simulator uses completely different mecha-
nism to generate traffic flow compared with some relatively advanced simulators at
present that also aim to mimic the very detailed part of vehicle movement [11] [42].
Models in these simulators also consider the effect of intersections, but vehicles be-
have unnaturally (e.g. stop for a prescribed amount of period) though they eventually
stop according to traffic rules. Besides, for traffic flows, these simulators usually use
probability models for vehicles to randomly choose a direction to go. In our case, we
can specify trip for each individual driver, hence there is no random component in
our simulation. In other words we control all the details of the test scenarios. As a
result, average speed is not as much valid a metric to measure our model. It is simply
insufficient to model realistic traffic flow with macroscopic features only.
Nevertheless, the difference could still be found in the traffic density at inter-
sections. Fig. 5.2 describes the vehicle density of Intelligent Driver Model and its
extension with Intersection Management. It is already a great progress in terms of
driver behavior coordination with traffic rules. From macroscopic view, it gives a per-
fect representation for realistic traffic flows, from microscopic point of view however,
it has several defects:
29
Fig. 5.1. Traffic Density at Intersection
(a) (b)
Fig. 5.2. Vehicular density [11]: (a) IDM model (b) IDM-IM traffic light model
(1) The main problem is that the peak value appears at the center of intersections
which is basically unrealistic, because vehicles only stops at the edges of intersections
so that traffic density around this area is higher.
(2) The vehicle density on the road apparently should be lower but not completely
homogeneous.
Fig. 5.1 gives an example of vehicular density at a intersection in our simulation
scenario with pre-defined traffic.
The Microscopic aspect of a traffic model is intended to describe the kinetic char-
acteristics of a particular vehicle. These kinetic factors include acceleration, decel-
eration, and drivers reaction towards surrounding objects. The micro feature of a
30
traffic model has significant impact on the realism of simulations, as the mobility
pattern of nodes could affect the route discovery process, link living time, and thus
fundamentally determines if a network could be feasibly supported by a specific rout-
ing protocol. For instance, link breaks could take place frequently when nodes are
highly mobile. The Intelligent Driver Model (IDM) [12] characterized the dynamics
of a particular vehicle as a function of multiple elements including the behavior of the
vehicles in front. As a time-continuous car following model, the IDM calculates the
instantaneous velocity and acceleration of a vehicle using the following differential
equations:
.xα =
dxαdt
= vα (5.1)
.vα =
dvαdt
= a(1− (vαv0
)4 − (s∗(vα,∆vα)
sα)2) (5.2)
s∗(vα,∆vα) = s0 + vαT +vα∆vα
2√ab
(5.3)
In these equations, xα and vα are the position and velocity of vehicle α , sα and
∆vα are defined as the current distance and velocity difference with vehicle α− 1 in
front, v0 stands for the ideal speed the driver desires to drive at, T denotes the safe
headway time to the front vehicle and thus s∗ is calculated as comfortable distance
to keep based on the minimum distance s0 between two vehicles in a traffic jam.
The IDM is a pure car following model that considers only the adjustment of
vehicle speed on the basis of parameters of the front vehicle, this works well with
roads with single lane, while in most scenarios (i.e. urban or freeway) roads are
more likely to be multi-lanes which makes the lane change possible. Because of this
limitation, the IDM was extended with Lane Changes (IDM-LC) and Intersection
Management (IDM-IM) in [11] to solve the issues explained above.
However, models like PTSM, TLM, SSM, IDM-IM, and IDM-LC mentioned above
did not give a detailed description of the intersection control. In this thesis, we focus
31
on the simulation of interactions between vehicles going straight and those turning
left at intersections. In connection to these factors, we use the driving simulator
together with the HyperDrive Suite [15] as our mobility generator for its integration
of the following components:
Intersection Control : Traffic signs could lead to special vehicle movement features
at intersections, because vehicles tend to decelerate and then stop before crossing the
intersection given corresponding traffic signals. As a result, the node density at inter-
sections should be much higher and thus provides longer living time for links among
this area. In our model, vehicles will slow down with the prescribed deceleration value
when approaching the intersection. Fig. 5.3 shows an example of four cars trying to
decelerate orderly in the same lane getting close to an intersection. This shows sharp
contrast with mobility models used in other traffic simulators as depicted in Fig. 5.4:
the intersection is either ignored or the vehicle did not entirely stop.
The status of traffic signal lights could be manually changed by the Intersection-
SetSignalState command with additional augments specifying the corresponding type,
direction and cycle of signal lights to configure.
Fig. 5.3. Speed variation of 4 vehicles in a row appoaching red light
In attempt to simulate the traffic lights at intersections, we coordinated the signal
controlling traffic flows from different directions. It should be noticed that left turn
32
Fig. 5.4. Speed variation of vehicles approaching intersections in other models [29]
signal does not necessarily turn green in each cycle, instead it depends on if there
are vehicles stopping at left turn only lanes. We have created some sample scenarios
to testify the correctness of the interactions between vehicle movements and traffic
signals. Fig. 5.5 illustrates the variation of all 8 traffic lights status from 4 directions
with time. In this scenario, we defined some traffic flows crossing the intersection
with only one direction (facing west) with vehicles turning left. This explains why
only the left turn signal facing west turns green from 40s to around 60s. Fig. 5.6
further describes drivers response approaching this intersection considering the right
of way with appropriate acceleration and deceleration actions taken. We may notice
that the green arrow gives left turn vehicle the priority.
With these examples we are confident that the logic of driver behavior is strictly
according to corresponding traffic signs. By observation of the speed curve showed
above, the speed variation is generally natural with proper acceleration and deceler-
ation.
Driving behavior : To better understand the real-world traffic, we need to consider
both individual driving behavior and interactions between them. In our work, by
default each vehicle placed in the scenario is configured to obey traffic rules such as
speed limit and take necessary actions with safety as the first priority.
However, it is though unreflective to assume that all vehicles move with the same
pattern as some aggressive driver could take more offensive actions while driving such
33
Fig. 5.5. Traffic light signal transition over time
Fig. 5.6. Evolution of vehicle speed from opposite directions whereleft-turn vehicle has the right of way
as overtake or even traffic sign violation. Regarding this, we change the following
characters of driving behavior to define different levels of aggressive driving:
• By EntityChangeRoadwaySpeed we increase the speed limit for some of the
vehicles, and they will try to overtake when possible.
• Considering some drivers choose to run yellow lights, the EntityChangeYellow-
LightGoTime is in charge of configuring the threshold time deciding whether the
specified vehicle will drive through a yellow light. If the vehicle is specified time or
34
less away from an intersection when the traffic signal turns yellow, then it chooses to
cross, otherwise the vehicle will stop as usual.
• In car-following mode, EntitySetHeadWay is used to determine the time interval
between the target vehicle and the entity in front of it. Additional augments Any-
oneInLane and UseLanePosition specify how the named entity will determine which
entity is in front of it. The option AnyoneInLane, which is the default behavior, will
cause the entity to stay behind anyone in the same lane regardless where in the lane
the entities are positioned. The other option, UseLanePosition will cause the named
entity to ignore entities that are not directly in front of it based on the lane offsets
and vehicle widths.
• When turning left without the right of way, the EntitySetLeftTurnGap makes
vehicles to wait for a gap in seconds of the specified size or larger before making a
turn.
• The EntityIgnoreIntersectionControl command causes the named vehicle to ig-
nore all intersection controls, such as traffic lights and stop signs. Neither will the
vehicle check for other vehicles entering the intersection.
5.1.2 Architecture
In this section, we primarily introduce our method to create scenario and trace ve-
hicle movement. Our simulation platform is composed of two parts: Driving simulator
as our mobility generator and NS-2 as network simulator. The general architecture is
shown in Fig. 5.7 The driving simulator with HyperDrive Suite is implemented with
Tool Command Language (TCL). The construction of scenarios includes the following
procedures:
• Authoring: The user defines the virtual world and all components that will
enable the driving scenarios. This includes specification of the roadway network,
traffic control devices, and events to be encountered as the scenario going on. Scripts
35
Fig. 5.7. Architecture of proposed simulation platform
are developed to control the logic for how the entities and scenario elements combine
to produce a realistic and interactive driving environment.
• Running: The driving simulator as a subject vehicle also participates in each run
of scenarios. During this session, the predefined data will be collected in real time.
In our case, we are more interested in the vehicle position and speed information.
• Reviewing: Our traffic simulator supports a powerful data recording system.
Multiple elements including vehicle speed, heading, acceleration, lane number, lane
offset, and position could be recorded in real time with specified frequency. After the
scenario ends, all the selected elements will be recorded in the data file. In our case
we are more interested in the vehicle trace, thus we choose the X, Y coordinates of
each vehicle in the simulation as output in 30Hz frequency.
The completion of a test scenario includes the following steps:
1) Creating Road Topology:
We build up the road topology by choosing tiles provided by the map editor shown
in Fig. 5.8 and then putting them together. A roadway tile is a square area with the
street layout inside. The library of HyperDrive Suite contains over 400 tiles, each with
36
Fig. 5.8. Map editor of HyperDrive
varying environment cultures (rural, urban, freeway, etc.) and road types defined by
number of lanes and traffic signals at intersections.
2) Encountered Events and Vehicle Movement:
The concept Trigger is used to define events taking place in scenarios in terms
of time and space. For instance, when specified vehicle reaches certain location then
events such as a pedestrian crossing street or traffic signal altering could be activated.
(a) (b)
Fig. 5.9. Visualization of traffic scenario and network animator withmovement traces imported: (a) HyperDrive. (b) NAM.
37
As with vehicle movement, we can manually place vehicles at any position on the
roadway as start point or use trigger to specify the time and location for a vehicle to
appear. We then use corresponding commands to define their behavior. In particular
we can specify the destinations or even the paths they will take.
5.2 Simulation
Our experiment is based on ns version 2.34. We evaluated the performance of Ad-
hoc On-Demand Distance Vector Routing (AODV), Dynamic Source Routing (DSR)
and Optimized Link State Routing (OLSR) protocols. For mobility trace, we collect
position data of all 80 vehicles in 30Hz frequency to ensure the replication of vehicle
traces and then use MATLAB to convert them into NS-2 readable format. A snap
shot of the animation of our simulation is given in Fig. 5.9 (b) using network animator
(NAM). We used the Two-ray Ground Reflection model as our radio wave propagation
model. The Two-ray Ground Reflection model [45] considers possible reflection via
ground and predicts the received power Pr at distance d with the following equation:
Pr =PtGtGrh
2th
2r
d2L(5.4)
Where ht and hr stand for the heights of the transmit and receive antennas in
meters respectively.
For each simulation with different number of traffic sources, we generate random
numbers as nodes IDs for traffic sources and destinations. However for different
routing protocols we use the same set of generated traffic sources for equilibrium
purpose. Table 5.1 gives a summarization of some other parameters we set for our
simulations.
We tested 80 vehicles with fixed initial positions and destinations crossing 4 neigh-
bor intersections controlled by traffic lights in a 2000 by 2000 area. The intersections
are equally distributed with 600 meters distance to each other. Most of the nodes in
the network are set to obey the speed limit, with other driving behaviors (left turn
38
Table 5.1Wireless simulation parameters
Network Simulator NS-2.34
Routing protocols AODV, DSR, OLSR
Simulation time 80s
Simulation area 2000m 2000m
Number of nodes 80
Number of traffic sources 5,10,15,20,25,30 (Randomly selected)
Data type UDP/ Constant byte rate
Packet size 512 bytes
Send Rate 4 packets/s
MAC IEEE 802.11
Transmission range 250m
Propagation Model Two-ray Ground
Table 5.2Traffic scenario parameters
Speed Limit 35 mlies/hour
Vehicle Yellow lights Go Time 1.5s
Vehicle Safe Headway 2s
Vehicle Left Turn Gap 7s
Vehicle Lane Change Headway 3s
Maximum Acceleration/Deceleration 3m/s2
Intersection Traffic Light Cycle 65s
Yellow light interval 3s
Vehicle Yellow lights Go Time 1.5s
39
gap, yellow light rushing time, headway, etc.) specified in Table 5.2. We randomly
picked 10 out of 80 vehicles with modified driving parameters to simulate aggressive
drivers.
40
6. RESULT ANALYSIS
6.1 Evaluation metrics
In order to compare and analyze the performance of routing protocols, we use
Packet Delivery Ratio (PDR), Average Packet Delay and end-to-end Throughput as
our evaluation metrics.
Fig. 6.1 shows the PDR values of AODV, OLSR, and DSR against number of CBR
sources. As we can observe, the general trend of PDR of all protocols is falling as the
number of traffic sources increases. This is because of the exponentially increasing
routing overhead with larger network scale generated by flooding-based routing pro-
tocols. Overall, the AODV and DSR perform significantly better than OLSR, which
suggests reactive routing protocols fit VANET better in terms of packet delivery. The
Average Packets Delay depicted in Fig. 6.2 displays opposite trend: OLSR is able to
deliver packets in less than 0.1s. AODV and DSR show conspicuous increment of
packet delay as DSR jumps up to 1.2s average delay (10 times more than that of
OLSR) as the number of CBR sources reaches 30, which is completely outperformed
by OLSR. This difference could be explained by the mechanisms employed in reactive
and proactive routing: Proactive routing updates routing table in real time, thus
upon sending a message the node already knows the paths to destinations, while in
contrast in reactive routing nodes need to send routing request before sending packets
which increases the delay.
6.2 Comparison with RWP model
To illustrate the necessity of using such highly detailed movement trace, we com-
pare our simulation result against the performance simulated under other common-
41
Fig. 6.1. Packet Delivery Ratio with varying number of CBR
Fig. 6.2. Average Packet Delay with varying number of CBR
ly used mobility environment. Using the completely same simulation area and ini-
tial node position, we also conducted some simulations based on the random way
42
point model. For the purpose of identifying the influence by mobility model only,
we use the random CBR traffic generator to produce exactly identical set of traffic
source/destination pairs.
For random waypoint model, since the node traces are with high randomness, we
conducted multiple simulations for each number of CBR sources. (Each point stands
for a mean of 6 experiments). The error bar has indicated the scale of variance. But
for the driving scenarios in the driving simulator, vehicles tend to move with certain
logic (e.g. interaction with drivers and traffic rules), the traces are only with slight
difference each run, hence the error is not that obvious.
Fig. 6.3 shows the packet delivery ratio of OLSR with varying number of CBR
sources from 5 to 30. In general, the packet delivery ratio decreases with the increment
of traffic sources, which indicates the network is gradually becoming congested as the
routing overhead increases drastically. As we can notice, the packet delivery ratio is
higher in RWP model. This could be explained by the characteristics of this model:
Nodes with fully random and free movement pattern tend to gather and form stable
link with each other thus providing a better environment for data delivery. However
this does not indicate that sophisticated movement pattern always lead to lower
performance compared with random models, the most important revelation here is
traffic pattern does have impact on performance evaluation. We have discussed the
unclearness of relation between mobility model and simulation result in previous
chapters, the point is that we proved the existence of such relation. Hence, the result
from Fig. 6.3 has well stated our purpose of using a highly detailed traffic simulator.
6.3 Analysis of throughput at intersections
As mentioned in previous sections, the particular vehicle movement features at
intersections could lead to better link state in terms of longer live time, as nodes tend
to form a relatively stable network topology. To further study and confirm the impact
by intersections, we have deliberately extracted some nodes around intersections from
43
Fig. 6.3. Comparison of Driving Simulator traces and Random Waypoint
our simulation, and compared their throughput when stopping at and leaving the
intersection. Fig. 6.4 (a) depicts the transferred throughput at node 5 which stays at
the intersection centered near coordinate (700,700) waiting for a chance to take left
turn at the beginning of each scenario. The throughput is averaged from all testing
scenarios. It is quite obvious that the throughput drops intensively after 40s. Then we
map the throughput data into spatial distribution using the position record of node 5.
The throughput versus position relationship is given in Fig. 6.4 (b). We find out when
placed at intersection, nodes are more likely to be involved as intermediate packet
relay to help complete the routing. The sharp comparison confirms our expectation.
To support that this phenomenon does not episodically appear at this particular node,
Fig. 6.5 (a) together with Fig. 6.5 (b) describes the average transferred throughputs
from node 71 in all scenarios. The graph shows similar trends: As we may examine,
the throughput reaches maximum at intersection centered at (700, 1300) and begins
to head downward after turning right. Upon reaching the next intersection, the
throughput regrows.
44
(a)
(b)
Fig. 6.4. Average throughput transferred at Node 5. (a) Time distri-bution (b) spatial distribution.
45
(a)
(b)
Fig. 6.5. Average throughput transferred at Node 71. (a) Time dis-tribution (b) spatial distribution.
46
7. CONCLUSION AND FUTURE WORK
A highly realistic VANETs simulation requires lots of factors to be accounted
to strengthen the cogency of the simulation result. These factors mainly fall into
two aspects, mobility model and propagation model. This work presents a traffic
simulator that integrates various benefits from the state-of-the-art of road traffic
simulations. By looking into the very details of movement patterns generated, we
thereby outlined its accuracy and fidelity of simulating realistic traffic flow especially
regarding intersection control.
We introduced our initial efforts of constructing a framework based on the combi-
nation of driving simulator and NS-2 to evaluate the performance of several routing
protocols. By analysis of the simulation results, we give a summarization of conclu-
sions to draw as following:
• Different routing protocols display their own characteristics when applied in
vehicular environment. Specifically, reactive routing strategies can provide relatively
higher delivery ratio compared with proactive routings. On the other hand, proac-
tive routing protocols manifest prominence in terms of packet delay. The perfor-
mance of reactive and proactive routing has presented us a trade-off issue, as in some
safety-related applications both the precision and swiftness of packet transmission are
desired. It is still premature to say which routing protocol best fits the VANETs.
• By studying the performance metrics of individual node near intersections, we
find out vehicles waiting at intersections provide more capability of transferring data
as a relay node. This might be helpful in the design of position-based routing protocols
One of the restrictions imposed by our mobility trace is the nature of being off-line
generated. We are able to evaluate the impact of mobility on the routing protocols
but not vice versa. However, our simulation still remains grounded because only the
bidirectional communication between traffic generator and network simulator is only
47
needed when applications (such as warning signal propagation) are accounted. In the
future we plan to build up bi-directional coupled simulator to comprehensively study
the interactive relation between network protocol and mobility model.
LIST OF REFERENCES
48
LIST OF REFERENCES
[1] T. Camp, J. Boleng, and V. Davies, “A Survey of Mobility Models for Ad HocNetwork Research,” Wireless Communication and Mobile Computing (WCMC):Special issue on Mobile Ad Hoc Networking: Research, Trends and Applications,vol. 2, no. 5, pp. 483-502, September 2002.
[2] J. Y. Le Boudec and M. Vojnovic, “The Random Trip Model: Stability, StationaryRegime, and Perfect Simulation.” IEEE/ACM Transactions on Networking, Vol.14, No. 6, pp. 1153-1166, December 2006.
[3] J. Yoon, M. Liu and B. Noble, “A General Framework to Construct StationaryMobility Models for the Simulation of Mobile Networks,” IEEE Transactions onMobile Computing, Vol. 5, No. 7, pp. 860-871, July 2006.
[4] F. Bai, N. Sadagopan, and A. Helmy. “IMPORTANT: A framework to system-atically analyze the Impact of Mobility on Performance of Routing protocols forAdhoc Networks,” INFOCOM 2003. Twenty-Second Annual Joint Conference ofthe IEEE Computer and Communications. IEEE Societies. Vol. 2. pp. 825-835,March 2003.
[5] A. Mahajan, N. Potnis, and K. Gopalan, “Urban mobility models for vanets.” inProc. 2nd IEEE International Workshop on Next Generation Wireless Networks.December 2006.
[6] A. Gainaru, C. Dobre, and V. Cristea, “A realistic mobility model based onsocial networks for the simulation of VANETs.” in Proc. IEEE 69th VehicularTechnology Conference, April 2009.
[7] M. Musolesi and C. Mascolo,“A community based mobility model for ad hocnetwork research,” in Proc. of the ACM 2nd International Workshop on Multi-hop Ad Hoc Networks: from theory to reality, pp. 31-38, May 2006.
[8] D. Choffnes and F. Bustamante, “An integrated mobility and traffic model forvehicular wireless networks,” in Proc. 2nd ACM International Workshop on Ve-hicular Ad Hoc Networks, pp. 69-78, September 2005.
[9] F. Karnadi, Z. Mo, and K. Lan. “Rapid generation of realistic mobility models forVANET.” in Proc. IEEE Wireless Communications and Networking Conference,pp. 2506-2511, March 2007.
[10] The Network Simulator NS 2. http://www.isi.edu/nsnam/ns/index.html/. July,2013.
[11] M. Fiore, J. Harri, F. Filali, and C. Bonnet, “Vehicular mobility simulation forVANETs,” in Proc. IEEE 40th Simulation Symposium, pp. 26-28, March 2007.
49
[12] M. Trieber, A. Hennecke, D. Helbing, “Congested traffic states in empirical obser-vations and microscopic simulations”, Physical Review E, pp. 1805-1824, August2000.
[13] C. Bettstetter, “Smooth is better than sharp: a random mobility model forsimulation of wireless networks,” in Proc. 4th ACM international workshop onModeling, analysis and simulation of wireless and mobile systems, pp. 19-27, 2001.
[14] J. Kraaier, U. Killat, “The Random Waypoint City Model, User Distributionin a Street-Based Mobility Model for Wireless Network Simulations”, in Proc.3rd ACM international workshop on Wireless mobile applications and services onWLAN hotspots, pp. 100-103, September 2005.
[15] W. Hsu, K. Merchant, H. Shu, C. Hsu, and A. Helmy, “Weighted WaypointMobility Model and its Impact on Ad Hoc Networks”, in Proc. ACM MobileComputer Communications Review (MC2R), January 2005.
[16] A. Einstein, “Investigations on the Theory of the Brownian Motion, in CollectedPapers”, R. Furth, pp. 170-182, 1926.
[17] http://www.drivesafety.com/. July 2013.
[18] C. Perkins, E. Belding-Royer and S. Das, “Ad hoc On-Demand Distance Vector(AODV) Routing,” in proc. 2nd IEEE Workshop on Mobile Computing Systemsand Applications, pp. 90-100, February 1999.
[19] D.B. Johnson, D.A. Maltz, and B. Josh. “DSR: The dynamic source routingprotocol for multi-hop wireless ad hoc networks,” Ad hoc networking 5, pp. 139-172, 2001.
[20] T. Clausen, P. Jacquet, C. Laouiti, P. Minet, P. Muhlethaler, A. Qayyum, andL. Viennot. “Optimized link state routing protocol (OLSR),” IETF RFC 3626,2003.
[21] C. Sommer, Christoph and F. Dressler. “Progressing toward realistic mobilitymodels in VANET simulations.” Communications Magazine, IEEE46.11, pp. 132-137, 2008
[22] D. Jiang and L. Delgrossi. “IEEE 802.11 p: Towards an international standardfor wireless access in vehicular environments.”Vehicular Technology Conference,pp. 2036-2040, 2008.
[23] Center for Microcomputers in Transportation (McTrans), Traffic soft-ware integrated systemcorridor simulation (TSIS-CORSIM), available athttp://mctrans.ce.ufl.edu/featured/tsis/. July 2013.
[24] PTV America, VISSIM, available at http://www.ptvamerica.com/vissim.html.July 2013.
[25] Quadstone, Inc., The PARAMICS transportation modeling suite, available athttp:// www.paramicsonline.com/. July 2013.
[26] SUMO, available at http://sumo.sourceforge.net/. July 2013.
50
[27] D. Krajzewicz, M. Bonert and P. Wagner. “The open source traffic simulationpackage SUMO”, in RobotCup 2006 Infrastructure Simulation Competition, Bre-men, Germany, 2006.
[28] M. Sanchez and P. Manzoni. “A java-based ad hoc networks simulator”. In Pro-ceedings of the SCS Western Multiconference Web-based Simulation Track, pp.573-583, January 1999.
[29] S. Olariu, and M.C. Weigle, eds. Vehicular Networks: From Theory To Practice.CRC Press, 2010.
[30] J. Yoon, M. Liu, and B. Noble. “Random Waypoint Considered Harmful,” Proc.IEEE Infocom, pp. 1312-1321, April 2003.
[31] http://www.census.gov/geo/www/tiger/. July 2013.
[32] D. Kotz and T. Henderson. “CRAWDAD: A Community Resource for ArchivingWireless Data at Dartmouth”. IEEE Pervasive Computing, pp. 12-14, October-December 2005.
[33] A. Qayyum, L.Viennot and A. Laouiti. “Multipoint Relaying for Flooding Broad-cast Messages in Mobile Wireless Networks”, in Proc. of the Hawaii InternationalConference on System Sciences (HICSS 02), pp. 3866-3875, Big Island, HI, USA,2002.
[34] M. Garey and D. Johnson. “Computers and Intractability: A Guide to the The-ory of NP-Completeness”, W.H. Freeman, San Francisco, CA, 1979.
[35] J. Jubin and J.D. Tornow. “The DARPA Packet Radio Network Protocols”, inProc. of the IEEE, Vol. 75, No. 1, pp. 21-32, January 1987.
[36] F.K. Karnadi, Z.H. Mo, and K. Lan. “Rapid generation of realistic mobility mod-els for VANET.” Wireless Communications and Networking Conference, IEEE,pp. 2506-2511, 2007.
[37] A. Wegener, M. Piorkowski, M. Raya, H. Hellbrck, S. Fischer, and J.P. Hubaux.“TraCI: An Interface for Coupling Road Traffic and Network Simulators”, in Proc.11th Communications and Networking Simulation Symposium (CNS08), pp. 155-163, 2008.
[38] S.Y. Wang, C.L. Chou, Y.H. Chiu, Y.S. Y.S. Tzeng, M.S. Hsu, Y.W. Cheng,W.L. Liu, and T.W. Ho. “NCTUns 4.0: An integrated simulation platform forvehicular traffic, communication, and network researches”, in Proc. IEEE VTC-Fall, pp. 2081-2085, 2007.
[39] CanuMobiSim, available at http://canu.informatik.uni-stuttgart.de. July 2013.
[40] M. Fiore,“Mobility Models in Inter-Vehicle Communications Literature”, Tech-nical Report, November 2006.
[41] O.K. Tonguz, N. Wisitpongphan, J.S. Parikht, F. Bait,P. Mudaliget, and V.K.Sadekar, “On the Broadcast Storm Problem in Ad hoc Wireless Networks”, inProc. the IEEE International Conference on Broadband Communications, Net-works and Systems (BROADNETS), pp. 1-11, 2006.
51
[42] A. Mahajan. Urban mobility models for vehicular ad hoc networks. Mastersthesis, 2006.
[43] F.J. Martinez, C.K. Toh, J. Cano, C.T. Calafate and P. Manzoni. “A surveyand comparative study of simulators for vehicular ad hoc networks (VANETs).”Wireless Communications and Mobile Computing, pp. 813-828, November 2011.
[44] J.B. Kenney. “Dedicated short-range communications (DSRC) standards in theUnited States.” Proceedings of the IEEE 99.7, pp. 1162-1182, 2011.
[45] T.S. Rappaport. Wireless Communications, Principles and Practice. PrenticeHall, 1996.
[46] http://www.dot.gov/ June 2013.
[47] http://www.nhtsa.gov/. June 2013.
[48] E. Rensel, D. Lebo, B. Graves, K. Malarich, C. Yorks. “Traffic Incident Manage-ment: Cost Management and Cost Recovery”. No. FHWA-HOP-12-044, 2012.